Information recommendation method and device based on artificial intelligence, electronic equipment and storage medium

Abstract

This invention provides an information recommendation method, apparatus, electronic device, and computer-readable storage medium based on artificial intelligence. The method includes: traversing historical interaction behavior data of a target object to determine a sequence of historical interaction information within the historical interaction behavior data; performing time-series-based one-way encoding on the vectors of the historical interaction information sequence to obtain a time-series interest vector sequence of the target object; performing two-way decoding on the time-series interest vector sequence to obtain a sequence vector of information to be recommended corresponding to each recommendation position; mapping multiple sequence vectors of information to be recommended to obtain information to be recommended corresponding to each recommendation position; and generating a recommendation list based on the information to be recommended corresponding to each recommendation position. This invention enables the parallel output of multiple pieces of information to be recommended, improving the efficiency of information recommendation.

Description

Technical Field

[0001] This invention relates to artificial intelligence technology, and more particularly to an information recommendation method, apparatus, electronic device, and computer-readable storage medium based on artificial intelligence. Background Technology

[0002] Artificial Intelligence (AI) is a comprehensive technology within computer science that studies the design principles and implementation methods of various intelligent machines, enabling them to possess perception, reasoning, and decision-making capabilities. AI technology is a multidisciplinary field, encompassing a wide range of areas, including natural language processing and machine learning / deep learning. With technological advancements, AI will be applied in more fields and play an increasingly important role.

[0003] Recommender systems are one of the important applications of artificial intelligence. They can help users discover information that may be of interest to them in an information-overloaded environment and push the information to users who are interested in it.

[0004] While recommender systems in related technologies can suggest information that users may be interested in, the efficiency and accuracy of these recommendations need improvement. Summary of the Invention

[0005] This invention provides an information recommendation method, apparatus, electronic device, and computer-readable storage medium based on artificial intelligence, which can output multiple pieces of information to be recommended in parallel, thereby improving the efficiency of information recommendation.

[0006] The technical solution of this invention is implemented as follows:

[0007] This invention provides an information recommendation method based on artificial intelligence, comprising:

[0008] The historical interaction behavior data of the target object is traversed to determine the sequence of historical interaction information in the historical interaction behavior data.

[0009] The vector of the historical interaction information sequence is subjected to time-series-based one-way encoding to obtain the time-series interest vector sequence of the target object;

[0010] Based on the time-series interest vector sequence, bidirectional decoding is performed to obtain the recommendation information sequence vector corresponding to each recommendation position;

[0011] The multiple sequences of information to be recommended are mapped to obtain the information to be recommended corresponding to each recommendation position, and a recommendation list is generated based on the information to be recommended corresponding to each recommendation position.

[0012] In the above technical solution, the step of mapping multiple sequence vectors of information to be recommended to obtain the information to be recommended corresponding to each recommendation position includes:

[0013] For any of the recommended locations, perform the following processing:

[0014] The probability distribution of the information to be recommended at the recommended position is obtained by performing nonlinear mapping on the sequence vector of information to be recommended at the recommended position.

[0015] The information to be recommended corresponding to the highest probability in the probability distribution is determined as the information to be recommended corresponding to the recommendation position.

[0016] In the above technical solution, the step of traversing the historical interaction behavior data of the target object to determine the historical interaction information sequence in the historical interaction behavior data includes:

[0017] The historical interaction behavior data of the target object is traversed to obtain the timestamp of each historical interaction information in the historical interaction behavior data.

[0018] The timestamps of each historical interaction information are sorted in descending order. The historical interaction information corresponding to the first few timestamps are combined, and the combined result is used as the historical interaction information sequence in the historical interaction behavior data.

[0019] This invention provides an information recommendation device, comprising:

[0020] The determination module is used to traverse the historical interaction behavior data of the target object in order to determine the historical interaction information sequence in the historical interaction behavior data.

[0021] The encoding module is used to perform time-series-based unidirectional encoding processing on the vector of the historical interaction information sequence to obtain the time-series interest vector sequence of the target object;

[0022] The decoding module is used to perform bidirectional decoding processing based on the time-series interest vector sequence to obtain the recommendation information sequence vector corresponding to each recommendation position;

[0023] The mapping module is used to map multiple sequence vectors of information to be recommended to obtain information to be recommended corresponding to each recommendation position, and to generate a recommendation list based on the information to be recommended corresponding to each recommendation position.

[0024] In the above technical solution, the device further includes:

[0025] The processing module is used to perform high-dimensional vector encoding processing on the historical interaction information sequence to obtain a high-dimensional vector corresponding to the historical interaction information sequence;

[0026] The high-dimensional vector is subjected to low-dimensional vector encoding to obtain a low-dimensional vector corresponding to the historical interaction information sequence.

[0027] The low-dimensional vector is used as the vector of the historical interaction information sequence;

[0028] The dimension of the high-dimensional vector is greater than the dimension of the low-dimensional vector.

[0029] In the above technical solution, the encoding module is further used to divide the historical interaction information sequence based on the temporal sequence of each historical interaction information in the historical interaction information sequence to obtain multiple sub-information sequences included in the historical interaction information sequence;

[0030] Perform the following processing on any one of the plurality of sub-information sequences:

[0031] The vector of the sub-information sequence is processed by a one-way convolution of the encoder to obtain the temporal interest vector corresponding to the sub-information sequence;

[0032] The position of the temporal interest vector corresponds to the edge position of the sub-information sequence;

[0033] Multiple temporal interest vectors are combined, and the combined result is used as the temporal interest vector sequence of the target object.

[0034] In the above technical solution, the encoder includes multiple cascaded coding layers, and each of the multiple coding layers corresponds to a different one-way convolution operation; the coding module is also used to perform one-way convolution processing on the vector of the sub-information sequence through the first coding layer of the multiple cascaded coding layers.

[0035] The convolution result of the first coding layer is output to the subsequent cascaded coding layers, so that unidirectional convolution processing and convolution result output can continue in the subsequent cascaded coding layers until the last coding layer is output.

[0036] The convolution result output from the last coding layer is used as the temporal interest vector corresponding to the sub-information sequence.

[0037] In the above technical solution, the decoding module is further used to concatenate the temporal interest vector sequence and the standard vector to obtain a concatenated vector sequence;

[0038] The concatenated vector sequence is processed by bidirectional convolution using a decoder to obtain a sequence vector of information to be recommended corresponding to each recommendation position.

[0039] In the above technical solution, the decoding module is further used to divide the spliced ​​vector sequence to obtain multiple sub-spliced ​​sequences included in the spliced ​​vector sequence;

[0040] For any one of the plurality of sub-concatenation sequences, perform the following processing:

[0041] The sub-concatenated sequence is processed by bidirectional convolution through a decoder to obtain the recommended information sequence vector corresponding to the recommended position;

[0042] The recommended position corresponds to the center position of the sub-segmentation sequence.

[0043] In the above technical solution, the decoder includes multiple cascaded decoding layers, and each of the multiple decoding layers corresponds to a different bidirectional convolution operation; the decoding module is also used to perform bidirectional convolution processing on the sub-concatenated sequence through the first decoding layer of the multiple cascaded decoding layers.

[0044] The convolution result of the first decoding layer is output to the subsequent cascaded decoding layers, so that bidirectional convolution processing and convolution result output can continue in the subsequent cascaded decoding layers until the last decoding layer is output.

[0045] The convolution result output from the last decoding layer is used as the sequence vector of information to be recommended for the corresponding recommendation position.

[0046] In the above technical solution, the mapping module is further configured to perform the following processing for any of the recommended locations:

[0047] The probability distribution of the information to be recommended at the recommended position is obtained by performing nonlinear mapping on the sequence vector of information to be recommended at the recommended position.

[0048] The information to be recommended corresponding to the highest probability in the probability distribution is determined as the information to be recommended corresponding to the recommendation position.

[0049] In the above technical solution, the device further includes:

[0050] The optimization module is used to perform masking processing on the information to be recommended corresponding to each recommendation position based on the mask, so as to obtain a mask sequence;

[0051] The decoding module is also used to perform bidirectional decoding based on the mask sequence to obtain recommendation information sequence vectors corresponding to each recommendation position.

[0052] The recommendation information sequence vectors corresponding to each recommendation position are mapped to obtain the recommendation information corresponding to each recommendation position.

[0053] In the above technical solution, the optimization module is further used to sort the probabilities of each piece of information to be recommended corresponding to the recommendation position in ascending order, and determine the information to be recommended corresponding to the first part of the probabilities as the information to be masked;

[0054] The mask information in multiple pieces of information to be recommended is updated to a mask to obtain a mask sequence.

[0055] In the above technical solution, the decoding module is further used to combine the temporal interest vector sequence and the mask sequence to obtain a combined sequence;

[0056] The combined sequence is processed by bidirectional convolution by the decoder to obtain recommendation information sequence vectors corresponding to each recommendation position.

[0057] In the above technical solution, the decoding module is further used to divide the combined sequence to obtain multiple sub-combined sequences included in the combined sequence;

[0058] For any one of the plurality of sub-combination sequences, perform the following processing:

[0059] The sub-combination sequence is processed by bidirectional convolution through a decoder to obtain a recommendation information sequence vector corresponding to the recommendation position;

[0060] The recommended position corresponds to the center position of the sub-combination sequence.

[0061] In the above technical solution, the decoder includes multiple cascaded decoding layers, and each of the multiple decoding layers corresponds to a different bidirectional convolution operation; the decoding module is also used to perform bidirectional convolution processing on the sub-combined sequence through the first decoding layer of the multiple cascaded decoding layers.

[0062] The convolution result of the first decoding layer is output to the subsequent cascaded decoding layers, so that bidirectional convolution processing and convolution result output can continue in the subsequent cascaded decoding layers until the last decoding layer is output.

[0063] The convolution result output by the last decoding layer is used as the recommendation information sequence vector corresponding to the recommendation position.

[0064] In the above technical solution, the determining module is further used to traverse the historical interaction behavior data of the target object to obtain the timestamp of each historical interaction information in the historical interaction behavior data;

[0065] The timestamps of each historical interaction information are sorted in descending order. The historical interaction information corresponding to the first few timestamps are combined, and the combined result is used as the historical interaction information sequence in the historical interaction behavior data.

[0066] This invention provides an electronic device for information recommendation, the electronic device comprising:

[0067] Memory, used to store executable instructions;

[0068] The processor, when executing executable instructions stored in the memory, implements the information recommendation method based on artificial intelligence provided in this embodiment of the invention.

[0069] This invention provides a computer-readable storage medium storing executable instructions that, when executed by a processor, implement the artificial intelligence-based information recommendation method provided in this invention.

[0070] The embodiments of the present invention have the following beneficial effects:

[0071] By encoding and decoding historical interaction information sequences, multiple pieces of information to be recommended can be output in parallel within a short period of time, thereby improving the efficiency of information recommendation. In addition, by using one-way encoding and two-way decoding, historical interaction information from multiple time series can be fused to improve the accuracy of information recommendation. Attached Figure Description

[0072] Figure 1 This is a schematic diagram illustrating an application scenario of the recommendation system provided in an embodiment of the present invention;

[0073] Figure 2 This is a schematic diagram of the structure of an electronic device for information recommendation provided in an embodiment of the present invention;

[0074] Figures 3A-3D This is a flowchart illustrating the information recommendation method based on artificial intelligence provided in an embodiment of the present invention;

[0075] Figure 4 This is a schematic diagram of the information recommendation process provided in an embodiment of the present invention;

[0076] Figure 5 This is a schematic diagram of the encoder provided in an embodiment of the present invention;

[0077] Figure 6 This is a schematic diagram of the decoder provided in an embodiment of the present invention;

[0078] Figure 7 This is a schematic diagram of the news application interface provided in an embodiment of the present invention;

[0079] Figure 8This is a schematic diagram of a shopping application interface provided in an embodiment of the present invention;

[0080] Figure 9 This is a schematic diagram of the network architecture provided in an embodiment of the present invention;

[0081] Figure 10 This is a schematic diagram of the encoder provided in an embodiment of the present invention;

[0082] Figure 11 This is a schematic diagram of the decoder provided in an embodiment of the present invention;

[0083] Figure 12 This is a schematic diagram of the mask optimization iterator provided in an embodiment of the present invention. Detailed Implementation

[0084] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0085] In the following description, the terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first" and "second" may be interchanged in a specific order or sequence where permitted, so that the embodiments of the invention described herein can be implemented in an order other than that illustrated or described herein.

[0086] In the implementation of this application, the collection and processing of relevant data should strictly comply with the requirements of relevant laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of laws and regulations and the authorization of the personal information subject.

[0087] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0088] Before providing a further detailed description of the embodiments of the present invention, the nouns and terms involved in the embodiments of the present invention will be explained, and the nouns and terms involved in the embodiments of the present invention shall be interpreted as follows.

[0089] 1) Convolutional Neural Networks (CNNs): A class of feedforward neural networks (FNNs) that include convolutional computations and have a deep structure, CNNs are one of the representative algorithms of deep learning. CNNs possess representation learning capabilities, enabling them to perform shift-invariant classification of input images according to their hierarchical structure.

[0090] 2) Target object: The object currently using the recommendation system (e.g., real users or virtual users simulated by computer programs). For example, if real user A is currently using the news recommendation system to browse news, then real user A is the target object.

[0091] Among related technologies, time-series recommendation methods for information recommendation include methods based on recurrent neural networks (RNNs) and methods based on convolutional neural networks (CNNs). While RNNs show advantages in recommendation accuracy, RNN-based methods can only process input data one by one, which is very time-consuming. Convolutional neural networks, although advantageous in recommendation efficiency, have relatively low ability and efficiency in modeling long sequences.

[0092] To address the aforementioned issues, embodiments of the present invention provide an information recommendation method, apparatus, electronic device, and computer-readable storage medium based on artificial intelligence. These methods can output multiple pieces of information to be recommended in parallel, improving the efficiency of information recommendation. Furthermore, by employing unidirectional encoding and bidirectional decoding, they integrate historical interaction information from various time series, thereby enhancing the accuracy of information recommendation.

[0093] The AI-based information recommendation method provided in this invention can be implemented by a terminal / server alone; or it can be implemented collaboratively by a terminal and a server. For example, the terminal can independently implement the AI-based information recommendation method described below, or the terminal can send a sequence of historical interaction information to the server, and the server can execute the AI-based information recommendation method based on the received historical interaction information sequence and send a recommendation list to the terminal.

[0094] The electronic device for information recommendation provided in this invention can be various types of terminal devices or servers. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal can be a smartphone, tablet computer, laptop computer, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited herein.

[0095] Taking servers as an example, such as server clusters deployed in the cloud, AI as a Service (AIaaS) is offered to users. The AIaaS platform breaks down several common AI services and provides them as independent or packaged services in the cloud. This service model is similar to an AI-themed marketplace, where all users can access and use one or more AI services provided by the AIaaS platform through application programming interfaces.

[0096] For example, one type of AI cloud service could be an information recommendation service, where a cloud server encapsulates the information recommendation program provided in this embodiment of the invention. Users invoke the information recommendation service in the cloud service through a terminal (running a client, such as a news client or shopping client), causing the cloud-deployed server to invoke the encapsulated information recommendation program. Based on the target object's historical interaction information sequence, the program performs encoding and decoding operations to obtain the information to be recommended corresponding to each recommendation position. It then generates a recommendation list based on the information to be recommended corresponding to each recommendation position to respond to information recommendation requests. For example, for news applications, a series of operations are performed based on the target user's historical clicked news sequence to obtain a news recommendation list, enabling a quick response to news recommendation requests. The target user can continuously browse news that matches their interests. For shopping applications, a series of operations are performed based on the target user's historical clicked product sequence to obtain a product recommendation list, enabling a quick response to product recommendation requests and recommending products that match the target user's interests, thereby increasing the target user's desire to shop.

[0097] See Figure 1 , Figure 1 This is a schematic diagram of the application scenario of the recommendation system 10 provided in the embodiment of the present invention. The terminal 200 is connected to the server 100 through the network 300. The network 300 can be a wide area network or a local area network, or a combination of the two.

[0098] Terminal 200 (running a client, such as a news client, shopping client, etc.) can be used to obtain requests for information recommendations to target users. For example, when a target user opens a news application, the terminal automatically obtains a request for news recommendations to the target user.

[0099] In some embodiments, an information recommendation plugin can be embedded in the client running on the terminal to implement an AI-based information recommendation method locally on the client. For example, after the terminal 200 receives a request for information recommendation for a target user, it calls the information recommendation plugin to implement an AI-based information recommendation method. This method performs encoding and decoding operations based on the target user's historical interaction information sequence to obtain the information to be recommended corresponding to each recommendation position. A recommendation list is then generated based on the information to be recommended corresponding to each recommendation position to respond to the request for information recommendation for the target user. For example, in a news application, when the target user scrolls through a news page, a request for news recommendation for the target user is automatically obtained. Based on the target user's historical click sequence of news, a series of operations are performed to obtain a news recommendation list, quickly responding to the request for news recommendation for the target user. This allows the target user to continuously browse news that matches their interests.

[0100] In some embodiments, after the terminal 200 receives a request for information recommendation for a target user, it calls the information recommendation interface of the server 100 (which can be provided as a cloud service, i.e., an information recommendation service). The server 100 performs encoding and decoding operations based on the historical interaction information sequence of the target object to obtain the information to be recommended corresponding to each recommendation position, and generates a recommendation list based on the information to be recommended corresponding to each recommendation position to respond to the request for information recommendation for the target user. For example, in a shopping application, when the target user scrolls through a product page, a request for product recommendation for the target user is automatically obtained, and a series of operations are performed based on the target user's historical click sequence of products to obtain a product recommendation list, so as to quickly respond to the request for news recommendation for the target user, thereby recommending products that match the target user's interests and enhancing the target user's desire to shop.

[0101] The structure of the electronic device for information recommendation provided in the embodiments of the present invention is described below. See also... Figure 2 , Figure 2 This is a schematic diagram of the structure of an electronic device 500 for information recommendation provided in an embodiment of the present invention. The example given is that the electronic device 500 is a server. Figure 2The illustrated electronic device 500 for information recommendation includes at least one processor 510, a memory 550, at least one network interface 520, and a user interface 530. The various components in the electronic device 500 are coupled together via a bus system 540. It is understood that the bus system 540 is used to implement communication between these components. In addition to a data bus, the bus system 540 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 2 The general labeled all buses as Bus System 540.

[0102] The processor 510 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0103] The memory 550 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), and the volatile memory may be random access memory (RAM). The memory 550 described in this embodiment is intended to include any suitable type of memory. The memory 550 may optionally include one or more storage devices physically located away from the processor 510.

[0104] In some embodiments, memory 550 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, as illustrated below.

[0105] Operating system 551 includes system programs for handling various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, driver layer, etc., for implementing various basic business functions and handling hardware-based tasks;

[0106] The network communication module 552 is used to reach other computing devices via one or more (wired or wireless) network interfaces 520, exemplary network interfaces 520 including: Bluetooth, WiFi, and Universal Serial Bus (USB), etc.

[0107] In some embodiments, the AI-based information recommendation device provided in this invention can be implemented in software, for example, as an information recommendation plugin in the terminal described above, or as an information recommendation service in the server described above. Of course, this is not the limitation; the AI-based information recommendation device provided in this invention can be provided in various software embodiments, including applications, software, software modules, scripts, or code.

[0108] Figure 2 An AI-based information recommendation device 555, stored in a memory 550, is shown. This device can be software in the form of programs and plugins, such as an information recommendation plugin, and includes a series of modules: a determining module 5551, an encoding module 5552, a decoding module 5553, a mapping module 5554, a processing module 5555, and an optimization module 5556. The determining module 5551, encoding module 5552, decoding module 5553, mapping module 5554, processing module 5555, and optimization module 5556 are used to implement the information recommendation function provided in this embodiment of the invention.

[0109] As mentioned above, the AI-based information recommendation method provided in this embodiment of the invention can be implemented by various types of electronic devices. See also Figure 3A , Figure 3A This is a flowchart illustrating the information recommendation method based on artificial intelligence provided in an embodiment of the present invention, combined with... Figure 3A The steps shown are explained.

[0110] In the following steps, the information to be recommended and the recommended information can be data such as text, images and text, and video. For example, for a search engine, the information to be recommended can be an answer in text form; for a news application, the information to be recommended can be a message in image and text form; and for a video application, the information to be recommended can be data in video form.

[0111] In step 101, the historical interaction behavior data of the target object is traversed to determine the historical interaction information sequence in the historical interaction behavior data.

[0112] As an example of acquiring historical interaction behavior data (i.e., behavioral data generated from the interaction between the target object and the electronic device, such as the target user's historical click behavior data, historical viewing data, etc.), when the target object scrolls through a page, the terminal automatically obtains a request for information recommendation for the target object and sends the request to the server. The server searches the behavior logs of the target object based on the request to obtain its historical interaction behavior data. After obtaining the target object's historical interaction behavior data, the server iterates through this data to obtain a sequence of historical interaction information for subsequent encoding and decoding operations. This sequence of historical interaction information includes multiple pieces of historical interaction information (i.e., data on the target object's interaction with the electronic device, such as historical click data, historical viewing data, etc.).

[0113] In some embodiments, the historical interaction behavior data of the target object is traversed to determine the historical interaction information sequence in the historical interaction behavior data, including: traversing the historical interaction behavior data of the target object to obtain the timestamp of each historical interaction information in the historical interaction behavior data; sorting the timestamps of each historical interaction information in descending order; combining the historical interaction information corresponding to the first sorted timestamps; and using the combination result as the historical interaction information sequence in the historical interaction behavior data.

[0114] For example, after the server obtains the historical interaction data of the target object, it iterates through this data to obtain the timestamp of each historical interaction. To retrieve the target object's most recent historical interaction information, the timestamps of each historical interaction are sorted in descending order to obtain the first N historical interaction messages, where N is a natural number set according to actual needs. The first N historical interaction messages are then combined in ascending order of their timestamps to form a historical interaction sequence. For example, the historical click information sequence could be [historical click information 1, historical click information 2, historical click information 3], where the timestamp of historical click information 1 is less than that of historical click information 2, and the timestamp of historical click information 2 is less than that of historical click information 3.

[0115] In step 102, the vector of the historical interaction information sequence is subjected to time-series-based one-way encoding to obtain the time-series interest vector sequence of the target object.

[0116] like Figure 4As shown, after the server obtains the historical interaction information sequence, the vector of the historical interaction information sequence can be processed by the encoder to perform time-series-based one-way encoding to obtain the time-series interest vector sequence of the target object (high-level sequence representation), thereby obtaining the interest representation of the target object, so as to obtain the recommendation information that matches the interests of the target object in the subsequent process.

[0117] In some embodiments, before performing time-based unidirectional encoding on the vector of the historical interaction information sequence to obtain the time-series interest vector sequence of the target object, the method further includes: performing high-dimensional vector encoding on the historical interaction information sequence to obtain a high-dimensional vector corresponding to the historical interaction information sequence; performing low-dimensional vector encoding on the high-dimensional vector to obtain a low-dimensional vector corresponding to the historical interaction information sequence, and using the low-dimensional vector as the vector of the historical interaction information sequence; wherein the dimension of the high-dimensional vector is greater than the dimension of the low-dimensional vector.

[0118] For example, after the server obtains the historical interaction information sequence, it needs to convert the historical interaction information sequence into word vectors. This can be done by first encoding the historical interaction information sequence into high-dimensional vectors to obtain high-dimensional vectors corresponding to the historical interaction information sequence. These high-dimensional vectors are used to accurately represent the historical interaction information and avoid missing information. Then, the high-dimensional vectors are encoded into low-dimensional vectors to obtain low-dimensional vectors corresponding to the historical interaction information sequence. These low-dimensional vectors are then used as vectors for the historical interaction information sequence, thus accurately representing the historical interaction information sequence through low-dimensional vectors and saving subsequent server computing resources.

[0119] See Figure 3B , Figure 3B This is an optional flowchart illustrating an artificial intelligence-based information recommendation method provided in an embodiment of the present invention. Figure 3B Show Figure 3A Step 102 in the process can be achieved through Figure 3B Steps 1021 to 1023 are implemented as follows: In step 1021, based on the temporal sequence of each historical interaction information in the historical interaction information sequence, the historical interaction information sequence is divided to obtain multiple sub-information sequences included in the historical interaction information sequence; In step 1022, the following processing is performed on any sub-information sequence among the multiple sub-information sequences: the vector of the sub-information sequence is subjected to unidirectional convolution processing by an encoder to obtain the temporal interest vector of the corresponding sub-information sequence; wherein, the position of the temporal interest vector corresponds to the edge position of the sub-information sequence; In step 1023, the multiple temporal interest vectors are combined, and the combined result is used as the temporal interest vector sequence of the target object.

[0120] like Figure 5As shown, firstly, based on the temporal order of each historical interaction information in the historical interaction information sequence, the historical interaction information sequence is divided into multiple sub-information sequences, where the timestamps of the historical interaction information in each sub-information sequence are arranged in ascending order. For example, Figure 5 In the shown sub-information sequence, the timestamp of node 0 is less than the timestamp of node 1. Then, for any sub-information sequence, the following processing is performed: the vector of the sub-information sequence is subjected to unidirectional convolution by the encoder to obtain the temporal interest vector of the corresponding sub-information sequence, where the position of the temporal interest vector corresponds to the edge position of the sub-information sequence, for example... Figure 5 The temporal interest vector (node ​​15) shown in the figure corresponds to the edge position in the sub-information sequence (i.e., Figure 5 The position of node 14 in the neutron information sequence could also be... Figure 5 (The position of node 0 in the neutron information sequence). Finally, all temporal interest vectors generated by all sub-information sequences are combined, and the combined result is used as the temporal interest vector sequence of the target object, for example, Figure 5 The temporal interest vectors shown in the figure (nodes 1 to 15 represent 15 temporal interest vectors) constitute a temporal interest vector sequence.

[0121] In some embodiments, the vector of the sub-information sequence is processed by a unidirectional convolution of the encoder to obtain the temporal interest vector of the corresponding sub-information sequence, including: performing unidirectional convolution processing on the vector of the sub-information sequence through the first coding layer of a plurality of cascaded coding layers; outputting the convolution result of the first coding layer to the subsequent cascaded coding layers, so as to continue to perform unidirectional convolution processing and output the convolution result in the subsequent cascaded coding layers, until the output is reached to the last coding layer; and using the convolution result output by the last coding layer as the temporal interest vector of the corresponding sub-information sequence.

[0122] Continuing with the above example, the encoder includes multiple cascaded coding layers, each corresponding to a different unidirectional convolution operation. To quickly capture long-term information of the target object, this embodiment employs dilated convolution to expand the receptive field. For example... Figure 5As shown, the encoder includes three coding layers. The first coding layer performs a one-way dilated convolution (dilation of 1) on the vector of the sub-information sequence to obtain the convolution result of the first coding layer. This result is then passed to the second coding layer. The second coding layer performs a one-way dilated convolution (dilation of 2) on the convolution result of the first coding layer to obtain the convolution result of the second coding layer. This result is then passed to the third coding layer. The third coding layer performs a one-way dilated convolution (dilation of 4) on the convolution result of the second coding layer to obtain the convolution result of the third coding layer. This convolution result of the third coding layer is used as the temporal interest vector of the corresponding sub-information sequence. Through hierarchical dilated convolution operations, the interest information of the target object can be extracted hierarchically, avoiding the omission of important interest information. This embodiment of the invention is not limited to dilated convolution; other convolution processing methods can also be used.

[0123] In step 103, bidirectional decoding is performed based on the temporal interest vector sequence to obtain the information sequence vector to be recommended corresponding to each recommendation position.

[0124] like Figure 4 As shown, after the server obtains the time-series interest vector sequence, it can perform bidirectional decoding on the time-series interest vector sequence to obtain the information to be recommended sequence vector corresponding to each recommendation position. This allows for the subsequent acquisition of information to be recommended that matches the target object's interests based on the information to be recommended sequence vector. Here, the recommendation position corresponds to the location of the information to be recommended in the recommendation list; for example, recommendation position 1 is the position of the first information to be recommended in the recommendation list, and recommendation position 2 is the position of the second information to be recommended in the recommendation list.

[0125] See Figure 3C , Figure 3C This is an optional flowchart illustrating an artificial intelligence-based information recommendation method provided in an embodiment of the present invention. Figure 3C Show Figure 3A Step 103 can be achieved through Figure 3C Steps 1031 to 1032 shown are implemented as follows: In step 1031, the temporal interest vector sequence and the standard vector are concatenated to obtain a concatenated vector sequence; In step 1032, the concatenated vector sequence is subjected to bidirectional convolution by a decoder to obtain the information sequence vector to be recommended corresponding to each recommendation position.

[0126] For example, after obtaining the temporal interest vector sequence, the server concatenates the temporal interest vector sequence with a standard vector (e.g., a 128-dimensional zero vector) to obtain a concatenated vector sequence. The decoder then performs bidirectional dilated convolution on the concatenated vector sequence to obtain the recommendation information sequence vector corresponding to each recommendation position.

[0127] In some embodiments, the concatenated vector sequence is subjected to bidirectional convolution processing by a decoder to obtain a sequence vector of information to be recommended corresponding to each recommendation position. This includes: dividing the concatenated vector sequence to obtain multiple sub-concatenated sequences; and performing the following processing on any one of the multiple sub-concatenated sequences: performing bidirectional convolution processing on the sub-concatenated sequence by a decoder to obtain a sequence vector of information to be recommended corresponding to the recommendation position; wherein the recommendation position corresponds to the center position of the sub-concatenated sequence.

[0128] like Figure 6 As shown, firstly, based on the temporal order of each concatenated vector in the concatenated vector sequence, the concatenated vector sequence is divided into multiple sub-concatenated sequences, where the timestamps of the concatenated vectors in each sub-concatenated sequence are arranged in ascending order. For example, Figure 6 In the shown sub-sequence, the timestamp of node 0 is less than the timestamp of node 1. Then, for any sub-sequence, the following processing is performed: the sub-sequence is subjected to bidirectional convolution by the decoder to obtain the recommendation information sequence vector corresponding to the recommendation position, where the recommendation position corresponds to the center position of the sub-sequence. For example... Figure 6 The sub-sequence shown is composed of nodes 4 to 10 and the recommended information sequence vector (node ​​8). The position of node 8 corresponds to the center position of the sub-sequence (i.e., Figure 6 (The position of node 7 in the neutron splicing sequence).

[0129] In some embodiments, the decoder includes multiple cascaded decoding layers, each corresponding to a different bidirectional convolution operation. The decoder performs bidirectional convolution processing on the sub-concatenated sequence to obtain a recommendation information sequence vector corresponding to the recommendation position. This includes: performing bidirectional convolution processing on the sub-concatenated sequence through the first decoding layer of the multiple cascaded decoding layers; outputting the convolution result of the first decoding layer to subsequent cascaded decoding layers to continue bidirectional convolution processing and outputting convolution results in subsequent cascaded decoding layers until the last decoding layer is output; and using the convolution result output by the last decoding layer as the recommendation information sequence vector corresponding to the recommendation position.

[0130] Continuing with the above example, the decoder includes multiple cascaded decoding layers, each corresponding to a different bidirectional convolution operation. To quickly capture future and past information of the target object, this embodiment employs bidirectional dilated convolution. For example... Figure 6As shown, the decoder includes two decoding layers. The first decoding layer performs a bidirectional dilated convolution (dilation factor of 1) on the concatenated sequence to obtain the convolution result of the first decoding layer. This result is then output to the second decoding layer. The second decoding layer performs a bidirectional dilated convolution (dilation factor of 2) on the convolution result of the first decoding layer to obtain the convolution result of the second decoding layer. This second convolution result is used as the vector of the recommendation information sequence for the corresponding recommendation position. Through hierarchical dilated convolution operations, decoding can be performed layer by layer to avoid missing important interest information. This embodiment of the invention is not limited to dilated convolution; other convolutional processing methods can also be used.

[0131] In step 104, multiple sequence vectors of information to be recommended are mapped to obtain the information to be recommended corresponding to each recommendation position, and a recommendation list is generated based on the information to be recommended corresponding to each recommendation position.

[0132] After the server obtains multiple sequence vectors of information to be recommended, it can map these multiple sequence vectors to obtain the information to be recommended corresponding to each recommendation position. Based on the information to be recommended corresponding to each recommendation position, a recommendation list is generated to respond to requests for information recommendations to the target object. According to the recommendation list, the information to be recommended is presented sequentially and continuously.

[0133] like Figure 7 As shown, for a news application, after the server obtains the recommendation list, it responds to the request to recommend news to the target user. As the target user scrolls through the news page, the news from the recommendation list is presented sequentially on the news page, so that the target user can continuously browse news that matches the target user's interests. For example, if the recommendation list includes [Recommendation position 1: News 1, Recommendation position 2: News 2, Recommendation position 3: News 3, Recommendation position 4: News 4], then news page 701 displays News 1, news page 702 displays News 2, news page 703 displays News 3, and news page 704 displays News 4.

[0134] like Figure 8 As shown, for a shopping application, after the server obtains the recommendation list, it responds to the request to recommend products to the target user. As the target user scrolls through the product page, the products in the recommendation list are displayed sequentially on the product page, so that the target user can continuously browse products that match the target user's interests. For example, if the recommendation list includes [Recommendation position 1: Product 1, Recommendation position 2: Product 2, Recommendation position 3: Product 3, Recommendation position 4: Product 4], then product page 801 displays Product 1, product page 802 displays Product 2, product page 803 displays Product 3, and product page 804 displays Product 4.

[0135] In some embodiments, mapping processing is performed on multiple information sequence vectors to be recommended to obtain information to be recommended corresponding to each recommendation position, including: performing the following processing on any recommendation position: performing nonlinear mapping processing on the information sequence vector to be recommended corresponding to the recommendation position to obtain the probability distribution of the information to be recommended at the recommendation position; and determining the information to be recommended corresponding to the highest probability in the probability distribution as the information to be recommended corresponding to the recommendation position.

[0136] For example, each recommendation position corresponds to a sequence vector of information to be recommended, and this sequence vector includes multiple vectors of information to be recommended. For any recommendation position, the following processing is performed: a non-linear mapping process (e.g., logistic regression function (softmax)) is applied to the sequence vector of information to be recommended corresponding to the recommendation position to obtain the probability distribution of the information to be recommended at that position. The information to be recommended corresponding to the highest probability in the probability distribution is then determined as the information to be recommended at that position. For example, the probability distribution of the information to be recommended at recommendation position 1 is [information to be recommended 1: 0.6, information to be recommended 2: 0.1, information to be recommended 3: 0.1, information to be recommended 4: 0.2], and the information to be recommended 1 corresponding to the highest probability in the probability distribution is determined as the information to be recommended at recommendation position 1.

[0137] See Figure 3D , Figure 3D This is an optional flowchart of an information recommendation method based on artificial intelligence provided in an embodiment of the present invention. In order to optimize the information to be recommended, a mask can be used to update the information to be recommended. Figure 3D The diagram also includes steps 105 to 107: In step 105, the information to be recommended corresponding to each recommendation position is masked based on the mask to obtain a mask sequence; in step 106, bidirectional decoding is performed based on the mask sequence to obtain a recommendation information sequence vector corresponding to each recommendation position; in step 107, the recommendation information sequence vector corresponding to each recommendation position is mapped to obtain the recommendation information corresponding to each recommendation position.

[0138] like Figure 4As shown, a mask optimization iterator is used to mask the information to be recommended at multiple recommendation positions to obtain a mask sequence. A decoder is then used to perform bidirectional decoding on the mask sequence to obtain a recommendation information sequence vector corresponding to each recommendation position. Finally, for any recommendation position, the following processing is performed: a non-linear mapping is applied to the recommendation information sequence vector corresponding to the recommendation position to obtain the probability distribution of the recommendation information at the recommendation position; the recommendation information corresponding to the highest probability in the probability distribution is determined as the recommendation information corresponding to the recommendation position, and a recommendation list is generated based on the recommendation information corresponding to each recommendation position to respond to the request for information recommendation for the target object. According to the recommendation list, the recommendation information is presented sequentially and continuously.

[0139] In some embodiments, the information to be recommended corresponding to each recommendation position is masked based on a mask to obtain a mask sequence, including: sorting the probability of each piece of information to be recommended corresponding to a recommendation position in ascending order, determining the information to be recommended corresponding to the first part of the probability as the information to be masked; updating the information to be masked among the multiple pieces of information to be recommended to a mask to obtain a mask sequence.

[0140] For example, in step 104, the probabilities of the information to be recommended corresponding to the recommendation position are obtained. The probabilities of each piece of information to be recommended corresponding to the recommendation position are sorted in ascending order. The information to be recommended at the top of the ranking has a relatively low probability, meaning its confidence and accuracy are relatively low. Therefore, it is necessary to optimize the information to be recommended at the top of the ranking. Thus, the information to be recommended corresponding to the top M probabilities is determined as the information to be masked, where M is a natural number set according to the actual application needs. The information to be masked among these multiple pieces of information to be recommended is then updated with a mask (i.e., a zero vector) to obtain a mask sequence.

[0141] In some embodiments, bidirectional decoding processing is performed based on the mask sequence to obtain recommendation information sequence vectors corresponding to each recommendation position, including: combining the temporal interest vector sequence and the mask sequence to obtain a combined sequence; and performing bidirectional convolution processing on the combined sequence through a decoder to obtain recommendation information sequence vectors corresponding to each recommendation position.

[0142] For example, after obtaining the mask sequence, the server combines the temporal interest vector sequence and the mask sequence to obtain a combined sequence, and then performs bidirectional dilated convolution on the combined sequence through the decoder to obtain recommendation information sequence vectors corresponding to each recommendation position.

[0143] In some embodiments, the combined sequence is subjected to bidirectional convolution processing by a decoder to obtain recommendation information sequence vectors corresponding to each recommendation position, including: dividing the combined sequence to obtain multiple sub-combined sequences included in the combined sequence; performing the following processing on any sub-combined sequence among the multiple sub-combined sequences: performing bidirectional convolution processing on the sub-combined sequence by a decoder to obtain recommendation information sequence vectors corresponding to the recommendation positions; wherein, the recommendation position corresponds to the center position of the sub-combined sequence.

[0144] Continuing with the example above, the combined sequence is first divided based on the temporal order of each vector, resulting in multiple sub-combined sequences. The timestamps of the vectors in each sub-combined sequence are arranged in ascending order. Then, for any sub-combined sequence, the following processing is performed: a decoder performs bidirectional convolution on the sub-combined sequence to obtain the recommendation information sequence vector corresponding to the recommendation position, where the recommendation position corresponds to the center position of the sub-combined sequence.

[0145] In some embodiments, the decoder includes multiple cascaded decoding layers, each corresponding to a different bidirectional convolution operation. The decoder performs bidirectional convolution processing on the sub-combined sequence to obtain a recommendation information sequence vector corresponding to the recommendation position. This includes: performing bidirectional convolution processing on the sub-combined sequence through the first decoding layer of the multiple cascaded decoding layers; outputting the convolution result of the first decoding layer to subsequent cascaded decoding layers to continue bidirectional convolution processing and outputting convolution results in subsequent cascaded decoding layers until the last decoding layer is output; and using the convolution result output by the last decoding layer as the recommendation information sequence vector corresponding to the recommendation position.

[0146] Continuing with the above example, the decoder includes multiple cascaded decoding layers, each corresponding to a different bidirectional convolution operation. For example, the decoder includes two decoding layers. The first decoding layer performs a bidirectional dilated convolution (dilation factor of 1) on the sub-combined sequence to obtain the convolution result of the first decoding layer. This convolution result is then output to the second decoding layer. The second decoding layer performs a bidirectional dilated convolution (dilation factor of 2) on the convolution result of the first decoding layer to obtain the convolution result of the second decoding layer. This convolution result is then used as the recommendation information sequence vector for the corresponding recommendation position. Through hierarchical dilated convolution operations, decoding processing can be performed hierarchically to avoid missing important interest information. This embodiment of the invention is not limited to dilated convolution; other convolution processing methods can also be used.

[0147] The following will describe an exemplary application of the embodiments of the present invention in a practical application scenario.

[0148] The embodiments of the present invention can be applied to various recommended application scenarios, such as Figure 1 As shown, terminal 200 connects to server 100 deployed in the cloud via network 300. After installing a news application on terminal 200 and obtaining a request for news recommendations for the target user, it calls the information recommendation interface of server 100. Server 100 performs a series of processes based on the request for news recommendations for the target user to obtain a news recommendation list, thereby quickly responding to the request for news recommendations for the target user. As a result, the target user can continuously browse news that matches the target user's interests.

[0149] Currently, temporal recommendation algorithms have been widely researched and applied in both industry and academia. This application scenario typically involves user interests changing over time, such as in short video recommendation scenarios; simultaneously, users may generate browsing history for hundreds of items (videos) within a few hours. In these short video applications, the recommendation system presents the user with an ordered list containing multiple videos at once. Users can then swipe through the videos in the list sequentially. This ordered video list recommendation in this scenario is called temporal recommendation. Good temporal recommendation methods can ensure a good user experience and have brought significant success to these platforms.

[0150] Temporal recommendation methods recommend items that match user preferences by studying the sequential relationships between user interactions (e.g., browsing or purchasing) within a sequence. While related recommendation methods (such as content-based and collaborative filtering) can only capture general user preferences without considering temporal information, temporal recommendation methods model user interactions as a dynamic sequence and then learn the high-level dependencies within that sequence to understand both long-term and short-term user preferences, thus achieving better recommendation performance.

[0151] For example, time-series recommendation algorithms in related technologies primarily focus on recommending an item to a user based on the items the user has already interacted with. Since these methods only consider the next item, if the recommended item is for the next K time points, the algorithm needs to generate it K times sequentially. This means the Kth recommended item strongly depends on the previous K-1 user clicks, and the (K-1)th recommended item strongly depends on the previous K-2 user clicks. This sequential approach, known as Next-K recommendation, is highly inefficient. Each recommendation request requires the server-side algorithm to recalculate and respond, resulting in slow response times and numerous server requests, leading to poor implementation results. Therefore, industrial applications often adopt a compromise strategy: when responding to the first recommendation request, all recommended items are directly sorted by score, and the top K items with the highest scores are used to form an ordered list of K recommended items—this is the top-K recommendation strategy—instead of generating recommended items for the next K time points through K requests. While this strategy saves a lot of time, experimental results show that the top-K recommendation strategy performs poorly in terms of recommendation accuracy and diversity when replacing the next-K strategy.

[0152] To address the aforementioned issues, this invention proposes an efficient Next-K recommendation algorithm (a temporal recommendation algorithm based on mask-optimized iterative dilated convolution), which simultaneously balances recommendation accuracy and efficiency, generating K future recommended items (recommendation information) in parallel. This proposed algorithm (an information recommendation method based on artificial intelligence) can accurately and efficiently generate an ordered recommendation list of K items. Specifically, this invention first captures user preferences through an encoder-decoder network structure and outputs an ordered recommendation list of K items in parallel. Deep dilated convolutional networks can learn the representation of long sequences relatively accurately with fewer parameters without losing information. Therefore, this invention uses a deep dilated convolutional network to implement the encoder, which can capture the sequence features of items that the user has interacted with and output a high-level sequence representation. Then, this invention uses a bidirectional deep dilated convolutional network to implement the decoder. Considering the high-level sequence representation and decoding position, the decoder can output an ordered list containing K items in parallel. Finally, this embodiment of the invention uses a mask-optimized iterator to iteratively optimize the generated ordered list.

[0153] Compared to time-series recommendation algorithms in related technologies (e.g., autoregressive generation algorithms, which generate K items sequentially and are executed K times to obtain K items; this method has relatively high accuracy but low efficiency; Top-K generation algorithms, which select the top K items with the highest scores from a list of recommended items to form a recommendation list; this method is efficient but has low accuracy), this invention proposes a non-autoregressive generation network that can effectively capture users' long and short-term time-series information and output a recommendation list containing K items in parallel and accurately in a short period of time.

[0154] The following is a detailed description of the dilated convolutional temporal recommendation algorithm based on mask optimization iteration proposed in this invention:

[0155] like Figure 9 As shown, the network architecture of this embodiment consists of a unidirectional dilated convolutional network, a bidirectional dilated convolutional network, and a mask optimization iterator. The unidirectional dilated convolutional network (encoder) captures both long-term and short-term information of the item sequence. Its input is the user's historical item interaction sequence, and its output is a high-level sequence representation. This high-level sequence representation is then input to the bidirectional dilated convolutional network (decoder), which outputs a recommendation list containing K items. To optimize the recommendation list output by the decoder, a mask optimization iterator selectively updates the decoder's recommendation list to generate an accurate recommendation list.

[0156] The network architecture of this invention includes: 1) an embedding layer; 2) an encoder layer; 3) a decoder layer; 4) a mapping layer (Softmax layer); and 5) a mask optimization iterator. The network architecture of this invention is described in detail below:

[0157] 1) Embedding layer

[0158] The Embedding layer is the first layer of the pre-trained algorithm model. It mainly maps the high-dimensional one-hot code to a low-dimensional embedding matrix. Each row of the matrix represents the embedding vector of an item, that is, the item vector (historical interaction information sequence).

[0159] 2) Encoder layer

[0160] The purpose of the encoder layer is to capture both long-term and short-term information of a sequence. Since the receptive field of a convolutional neural network (CNN) expands linearly with the number of convolutional layers, it is difficult for CNNs to capture long-term information from the user with just a few layers. Furthermore, more layers will cause the vanishing gradient problem and may significantly increase computation time. Therefore, this invention utilizes a unidirectional dilated neural network as the encoder, aiming to comprehensively and efficiently learn the patterns of a sequence. Its dilated convolutional network can expand the receptive field by introducing certain gaps in the convolutional kernels, meaning that its receptive field can be increased with fewer layers (i.e., parameters), thereby efficiently capturing both long-term and short-term information from the user.

[0161] like Figure 10 As shown, the receptive field of the first layer is 3, the receptive field of the second layer is 7, and the receptive field of the third layer is 15. In real-world long-sequence recommendation scenarios, for example, if a user watches 1000 short videos a day, then repeating... Figure 8 In the architecture, the dilated convolution dilation dilation value is set as follows: {1, 2, 4, ..., 128, 1, 2, 4, ..., 128, 1, 2, 4, ..., 128}. Finally, the encoder outputs a high-level sequence representation (a sequence of temporal interest vectors), and its encoder can be a NextItNet model, but is not limited to the NextItNet model.

[0162] 3) Decoder layer

[0163] The goal of the decoder is to decode the target sequence based on the high-level sequence representation captured by the encoder. The decoder in this embodiment employs a bidirectional dilated convolutional network. The convolution operation of the decoder is bidirectional, meaning that points in the upper layer can access the left and right points of the lower layer, for example, as shown below. Figure 11 As shown, node 8 in the high-level representation corresponds to node 7 in the high-level sequence representation. Node 8 in the high-level representation can access the nodes to the left of node 7 (nodes 4 to 6 in the high-level sequence representation) and the nodes to the right of node 7 (nodes 8 to 10 in the high-level sequence representation). Thus, information from the future and the past helps generate an accurate recommendation list.

[0164] The decoder's input is a concatenation of an all-zero vector and the high-level sequence representation output by the encoder. The dilated convolutional network can identify the positional information of different nodes, so no additional positional information needs to be added to the nodes. The decoder's output is a high-level representation (a sequence vector of recommended information) of the target position (recommended position) (there are K such vectors).

[0165] 4) Softmax layer

[0166] Based on the high-level representation output by the decoder, softmax processing is performed on all items included in each position to obtain the probability of all items included in each position. The item with the highest probability in any position is taken as the item for that position (the information to be recommended corresponding to the recommended position). The items of all positions constitute the item list.

[0167] 5) Mask-optimized iterator

[0168] Since decoding is based on a series of masked inputs, the resulting list of items may not be optimal. To address this issue, such as... Figure 12 As shown, this embodiment of the invention proposes a mask-optimized iterator to iteratively update the generated item list.

[0169] Specifically, this mask optimization iterator re-predicts low-confidence items (i.e., low-probability items) in the item list output by the decoder, iterating through the prediction process to ultimately predict a more accurate item list. Specifically, this iterator consists of two steps:

[0170] Step 1: Masking. Based on the probability of an item at each position obtained from softmax (i.e., the confidence level of that item), identify N items with low confidence levels and mask them. Here, N is a random positive integer between [1, K].

[0171] Step 2: Iteration (Generation) Input the vector without a mask and the masked item (all 0s) into the decoder, and generate a new list of items according to the process of 4) and 5).

[0172] The iteration stops after a preset number of iterations is reached, in order to generate the optimal list of items.

[0173] To verify the effectiveness of the method proposed in the embodiments of the present invention, four evaluation indicators were designed:

[0174] 1. Top-order: A high-level representation based on Next-1 item recommendations, directly recommending the Top-K items. When comparing with the labeled data (ground-truth) list, positional order is considered.

[0175] 2. Top-no-order: An advanced representation based on Next-1 item recommendations, directly recommending the Top-K items. When comparing with the ground-truth list, positional order is not considered; the intersection of the two lists is used.

[0176] 3. Next-order: A high-level representation based on the Next-K item recommendations, recommending a Next-K list. When comparing with the ground-truth list, the positional order is considered.

[0177] 4. Next-no-order: Based on a high-level representation of Next-K item recommendations, a Next-K list is recommended. When compared with the ground-truth list, the positional order is not considered; that is, the intersection of the two lists is taken.

[0178] For the classic recommendation system dataset (Movielens), the recommendation accuracy of this embodiment is shown in Table 1 (where the length of the observed sequence is 70 and the length of the sequence to be predicted is 30):

[0179] Table 1: Recommendation Accuracy

[0180]

[0181] The experimental results show that our method is significantly better than related technologies (e.g., NextItNet) in terms of recommendation accuracy.

[0182] To evaluate recommendation efficiency, the time (in seconds) for generating 128 lists in a CPU environment is compared between NextItNet and the method proposed in this embodiment, as shown in Table 2:

[0183] Table 2: Recommendation Efficiency

[0184]

[0185] The experimental results show that our method is significantly better than related technologies (e.g., NextItNet) in terms of recommendation efficiency.

[0186] In this embodiment of the invention, Adam can be used as the optimizer, the batch size is 128, the learning rate is 0.001, 90% of the training data is used as the training set, and the remainder is used as the test set.

[0187] In summary, the recommendation algorithm based on a dilated convolutional network with a unidirectional encoder and a bidirectional decoder proposed in this invention can effectively capture users' short-term and long-term temporal information and output a recommendation list containing K items with high accuracy in a short time. By using a mask optimization iterator, the generated recommendation list can be fine-tuned to reduce the repetition rate of items and improve the recommendation accuracy.

[0188] This concludes the description of the artificial intelligence-based information recommendation method provided by the embodiments of the present invention, using exemplary applications and implementations of the server provided in the embodiments of the present invention. The embodiments of the present invention also provide an information recommendation device. In practical applications, the functional modules in the information recommendation device can be collaboratively implemented using the hardware resources of electronic devices (such as terminal devices, servers, or server clusters), such as computing resources like processors, communication resources (such as those used to support various communication methods like optical fiber and cellular), and memory. Figure 2 An information recommendation device 555 stored in memory 550 is shown. It can be software in the form of programs and plug-ins, such as software modules designed in programming languages ​​such as C / C++ and Java, application software designed in programming languages ​​such as C / C++ and Java, or dedicated software modules, application programming interfaces, plug-ins, cloud services, etc. in large software systems. Examples of different implementation methods are given below.

[0189] Example 1: Information recommendation devices are mobile applications and modules.

[0190] The information recommendation device 555 in this embodiment of the invention can provide a software module designed using programming languages ​​such as C / C++ and Java, which is embedded in various mobile applications based on systems such as Android or iOS (stored as executable instructions in the storage medium of the mobile device and executed by the processor of the mobile device), thereby directly using the computing resources of the mobile device itself to complete the relevant information recommendation tasks, and periodically or irregularly transmitting the processing results to a remote server through various network communication methods, or saving them locally on the mobile device.

[0191] Example 2: The information recommendation device is a server application and platform.

[0192] The information recommendation device 555 in this embodiment of the invention can be a dedicated software module in application software or large software systems designed using programming languages ​​such as C / C++ and Java. It runs on the server side (stored in the server-side storage medium in the form of executable instructions and run by the server-side processor). The server uses its own computing resources to complete the relevant information recommendation tasks.

[0193] Embodiments of the present invention may also provide an information recommendation platform (for recommendation lists) for use by individuals, groups or organizations by mounting a customized, easy-to-interact web interface or other user interfaces on a distributed, parallel computing platform composed of multiple servers.

[0194] Example 3: The information recommendation device consists of server-side application programming interfaces (APIs) and plugins.

[0195] The information recommendation device 555 in this embodiment of the invention can be provided as a server-side API or plugin for users to call, to execute the artificial intelligence-based information recommendation method of this embodiment of the invention, and to be embedded in various applications.

[0196] Example 4: Information recommendation devices are mobile device client APIs and plugins.

[0197] The information recommendation device 555 in this embodiment of the invention can be provided as an API or plugin on a mobile device for users to call in order to execute the information recommendation method based on artificial intelligence in this embodiment of the invention.

[0198] Example 5: The pathology image processing device is an open service in the cloud.

[0199] The information recommendation device 555 in this embodiment of the invention can provide a cloud service for information recommendation developed for users, allowing individuals, groups or organizations to obtain recommendation lists.

[0200] The information recommendation device 555 includes a series of modules, including a determination module 5551, an encoding module 5552, a decoding module 5553, a mapping module 5554, a processing module 5555, and an optimization module 5556. The following description further illustrates the scheme by which the various modules in the information recommendation device 555 provided in this embodiment of the invention cooperate to achieve information recommendation.

[0201] The determination module 5551 is used to traverse the historical interaction behavior data of the target object to determine the historical interaction information sequence in the historical interaction behavior data; the encoding module 5552 is used to perform time-series-based one-way encoding processing on the vector of the historical interaction information sequence to obtain the time-series interest vector sequence of the target object; the decoding module 5553 is used to perform bidirectional decoding processing based on the time-series interest vector sequence to obtain the information sequence vector to be recommended corresponding to each recommendation position; the mapping module 5554 is used to map multiple information sequence vectors to be recommended to obtain the information to be recommended corresponding to each recommendation position, and generate a recommendation list based on the information to be recommended corresponding to each recommendation position.

[0202] In some embodiments, the information recommendation device 555 further includes: a processing module 5555, configured to perform high-dimensional vector encoding processing on the historical interaction information sequence to obtain a high-dimensional vector corresponding to the historical interaction information sequence; perform low-dimensional vector encoding processing on the high-dimensional vector to obtain a low-dimensional vector corresponding to the historical interaction information sequence, and use the low-dimensional vector as the vector of the historical interaction information sequence; wherein the dimension of the high-dimensional vector is greater than the dimension of the low-dimensional vector.

[0203] In some embodiments, the encoding module 5552 is further configured to divide the historical interaction information sequence based on the temporal sequence of each historical interaction information in the historical interaction information sequence to obtain multiple sub-information sequences included in the historical interaction information sequence; and to perform the following processing on any one of the multiple sub-information sequences: to perform unidirectional convolution processing on the vector of the sub-information sequence by an encoder to obtain a temporal interest vector corresponding to the sub-information sequence; wherein the position of the temporal interest vector corresponds to the edge position of the sub-information sequence; and to combine multiple temporal interest vectors to use the combination result as the temporal interest vector sequence of the target object.

[0204] In some embodiments, the encoder includes multiple cascaded coding layers, and each of the multiple coding layers corresponds to a different unidirectional convolution operation; the coding module 5552 is further configured to perform unidirectional convolution processing on the vector of the sub-information sequence through the first coding layer of the multiple cascaded coding layers; output the convolution result of the first coding layer to subsequent cascaded coding layers, so as to continue unidirectional convolution processing and convolution result output in the subsequent cascaded coding layers, until the output is reached to the last coding layer; and use the convolution result output by the last coding layer as the temporal interest vector corresponding to the sub-information sequence.

[0205] In some embodiments, the decoding module 5553 is further configured to concatenate the temporal interest vector sequence and the standard vector to obtain a concatenated vector sequence; and to perform bidirectional convolution processing on the concatenated vector sequence through the decoder to obtain a sequence vector of information to be recommended corresponding to each recommendation position.

[0206] In some embodiments, the decoding module 5553 is further configured to divide the spliced ​​vector sequence to obtain a plurality of sub-spliced ​​sequences included in the spliced ​​vector sequence; and to perform the following processing on any one of the plurality of sub-spliced ​​sequences: to perform bidirectional convolution processing on the sub-spliced ​​sequence through the decoder to obtain a recommendation information sequence vector corresponding to the recommendation position; wherein, the recommendation position corresponds to the center position of the sub-spliced ​​sequence.

[0207] In some embodiments, the decoder includes multiple cascaded decoding layers, and each of the multiple decoding layers corresponds to a different bidirectional convolution operation; the decoding module 5553 is further configured to perform bidirectional convolution processing on the sub-concatenated sequence through the first decoding layer of the multiple cascaded decoding layers; output the convolution result of the first decoding layer to the subsequent cascaded decoding layers, so as to continue bidirectional convolution processing and convolution result output in the subsequent cascaded decoding layers, until the output is reached to the last decoding layer; and use the convolution result output by the last decoding layer as the recommended information sequence vector corresponding to the recommended position.

[0208] In some embodiments, the mapping module 5554 is further configured to perform the following processing for any of the recommended positions: perform nonlinear mapping processing on the sequence vector of information to be recommended corresponding to the recommended position to obtain the probability distribution of the information to be recommended at the recommended position; and determine the information to be recommended corresponding to the highest probability in the probability distribution as the information to be recommended corresponding to the recommended position.

[0209] In some embodiments, the information recommendation device 555 further includes: an optimization module 5556, configured to perform masking processing on the information to be recommended corresponding to each recommendation position based on a mask to obtain a mask sequence; a decoding module 5553 is further configured to perform bidirectional decoding processing based on the mask sequence to obtain a recommendation information sequence vector corresponding to each recommendation position; and to perform mapping processing on the recommendation information sequence vector corresponding to each recommendation position to obtain recommendation information corresponding to each recommendation position.

[0210] In some embodiments, the optimization module 5556 is further configured to sort the probabilities of each piece of information to be recommended corresponding to the recommendation position in ascending order, determine the information to be recommended corresponding to the first part of the probabilities as information to be masked, and update the information to be masked among the multiple pieces of information to be recommended to a mask to obtain a mask sequence.

[0211] In some embodiments, the decoding module 5553 is further configured to combine the temporal interest vector sequence and the mask sequence to obtain a combined sequence; and to perform bidirectional convolution processing on the combined sequence through the decoder to obtain recommendation information sequence vectors corresponding to each recommendation position.

[0212] In some embodiments, the decoding module 5553 is further configured to divide the combined sequence to obtain a plurality of sub-combined sequences included in the combined sequence; and to perform the following processing on any one of the plurality of sub-combined sequences: to perform bidirectional convolution processing on the sub-combined sequence through the decoder to obtain a recommendation information sequence vector corresponding to the recommendation position; wherein, the recommendation position corresponds to the center position of the sub-combined sequence.

[0213] In some embodiments, the decoder includes multiple cascaded decoding layers, and each of the multiple decoding layers corresponds to a different bidirectional convolution operation; the decoding module 5553 is further configured to perform bidirectional convolution processing on the sub-combined sequence through the first decoding layer of the multiple cascaded decoding layers; output the convolution result of the first decoding layer to the subsequent cascaded decoding layers, so as to continue bidirectional convolution processing and convolution result output in the subsequent cascaded decoding layers, until the output is reached to the last decoding layer; and use the convolution result output by the last decoding layer as the recommendation information sequence vector corresponding to the recommendation position.

[0214] In some embodiments, the determining module 5551 is further configured to traverse the historical interaction behavior data of the target object to obtain the timestamp of each historical interaction information in the historical interaction behavior data; sort the timestamps of each historical interaction information in descending order; combine the historical interaction information corresponding to the first sorted timestamps; and use the combination result as the historical interaction information sequence in the historical interaction behavior data.

[0215] This invention provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the artificial intelligence-based information recommendation method described above in this invention.

[0216] This invention provides a computer-readable storage medium storing executable instructions. When these executable instructions are executed by a processor, they cause the processor to execute the artificial intelligence-based information recommendation method provided in this invention. For example, ... Figures 3A-3D The example shown is an information recommendation method based on artificial intelligence.

[0217] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0218] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0219] As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborating files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0220] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0221] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of the present invention are included within the scope of protection of the present invention.

Claims

1. An information recommendation method based on artificial intelligence, characterized in that, include: The historical interaction behavior data of the target object is traversed to determine the sequence of historical interaction information in the historical interaction behavior data. The vector of the historical interaction information sequence is subjected to time-series-based one-way encoding to obtain the time-series interest vector sequence of the target object; The temporal interest vector sequence and the standard vector are concatenated to obtain a concatenated vector sequence; the concatenated vector sequence is bidirectionally decoded to obtain a sequence vector of information to be recommended corresponding to each recommendation position, wherein each recommendation position corresponds to the center position of multiple sub-concatenated sequences included in the concatenated vector sequence; The multiple sequences of information to be recommended are mapped to obtain the information to be recommended corresponding to each recommendation position, and a recommendation list is generated based on the information to be recommended corresponding to each recommendation position.

2. The method according to claim 1, characterized in that, Before performing time-series-based unidirectional encoding on the vector of the historical interaction information sequence to obtain the time-series interest vector sequence of the target object, the method further includes: The historical interaction information sequence is subjected to high-dimensional vector encoding to obtain a high-dimensional vector corresponding to the historical interaction information sequence. The high-dimensional vector is subjected to low-dimensional vector encoding to obtain a low-dimensional vector corresponding to the historical interaction information sequence. The low-dimensional vector is used as the vector of the historical interaction information sequence; The dimension of the high-dimensional vector is greater than the dimension of the low-dimensional vector.

3. The method according to claim 1, characterized in that, The step of performing time-based unidirectional encoding on the vector of the historical interaction information sequence to obtain the time-series interest vector sequence of the target object includes: Based on the temporal sequence of each historical interaction information in the historical interaction information sequence, the historical interaction information sequence is divided to obtain multiple sub-information sequences included in the historical interaction information sequence; Perform the following processing on any one of the plurality of sub-information sequences: The vector of the sub-information sequence is processed by a one-way convolution of the encoder to obtain the temporal interest vector corresponding to the sub-information sequence; The position of the temporal interest vector corresponds to the edge position of the sub-information sequence; Multiple temporal interest vectors are combined, and the combined result is used as the temporal interest vector sequence of the target object.

4. The method according to claim 3, characterized in that, The encoder includes multiple cascaded coding layers, and each coding layer corresponds to a different unidirectional convolution operation; The step of performing unidirectional convolution processing on the vector of the sub-information sequence through an encoder to obtain the temporal interest vector corresponding to the sub-information sequence includes: The vector of the sub-information sequence is subjected to unidirectional convolution processing by the first coding layer of the multiple cascaded coding layers. The convolution result of the first coding layer is output to the subsequent cascaded coding layers, so that unidirectional convolution processing and convolution result output can continue in the subsequent cascaded coding layers until the last coding layer is output. The convolution result output from the last coding layer is used as the temporal interest vector corresponding to the sub-information sequence.

5. The method according to claim 1, characterized in that, The step of performing bidirectional decoding on the concatenated vector sequence to obtain the recommendation information sequence vector corresponding to each recommendation position includes: The concatenated vector sequence is processed by bidirectional convolution using a decoder to obtain a sequence vector of information to be recommended corresponding to each recommendation position.

6. The method according to claim 5, characterized in that, The step of performing bidirectional convolution processing on the concatenated vector sequence through a decoder to obtain the recommendation information sequence vector corresponding to each recommendation position includes: The spliced ​​vector sequence is divided to obtain the plurality of sub-spliced ​​sequences included in the spliced ​​vector sequence; For any one of the plurality of sub-concatenation sequences, perform the following processing: The sub-concatenated sequence is processed by bidirectional convolution using a decoder to obtain the recommended information sequence vector corresponding to the recommended position.

7. The method according to claim 6, characterized in that, The decoder includes multiple cascaded decoding layers, and each decoding layer corresponds to a different bidirectional convolution operation; The step of performing bidirectional convolution processing on the sub-concatenated sequence through a decoder to obtain the recommendation information sequence vector corresponding to the recommendation position includes: The sub-concatenated sequence is subjected to bidirectional convolution processing by the first decoding layer of the multiple cascaded decoding layers. The convolution result of the first decoding layer is output to the subsequent cascaded decoding layers, so that bidirectional convolution processing and convolution result output can continue in the subsequent cascaded decoding layers until the last decoding layer is output. The convolution result output from the last decoding layer is used as the sequence vector of information to be recommended for the corresponding recommendation position.

8. The method according to claim 1, characterized in that, After mapping the multiple sequences of information to be recommended, the process further includes: Based on the mask, the information to be recommended corresponding to each recommendation position is masked to obtain a mask sequence; Bidirectional decoding is performed based on the mask sequence to obtain recommendation information sequence vectors corresponding to each recommendation position; The recommendation information sequence vectors corresponding to each recommendation position are mapped to obtain the recommendation information corresponding to each recommendation position.

9. The method according to claim 8, characterized in that, The process of masking the information to be recommended at each recommendation position to obtain a mask sequence includes: Sort the probabilities of each piece of information to be recommended corresponding to the recommendation position in ascending order, and determine the information to be recommended corresponding to the first part of the probabilities as the information to be masked. The mask information in multiple pieces of information to be recommended is updated to a mask to obtain a mask sequence.

10. The method according to claim 8, characterized in that, The bidirectional decoding process based on the mask sequence to obtain recommendation information sequence vectors corresponding to each recommendation position includes: The temporal interest vector sequence and the mask sequence are combined to obtain a combined sequence. The combined sequence is processed by bidirectional convolution by the decoder to obtain recommendation information sequence vectors corresponding to each recommendation position.

11. The method according to claim 10, characterized in that, The step of performing bidirectional convolution processing on the combined sequence through a decoder to obtain recommendation information sequence vectors corresponding to each recommendation position includes: The combined sequence is divided to obtain multiple sub-combined sequences included in the combined sequence; For any one of the plurality of sub-combination sequences, perform the following processing: The sub-combination sequence is processed by bidirectional convolution through a decoder to obtain a recommendation information sequence vector corresponding to the recommendation position; The recommended position corresponds to the center position of the sub-combination sequence.

12. The method according to claim 11, characterized in that, The decoder includes multiple cascaded decoding layers, and each decoding layer corresponds to a different bidirectional convolution operation; The step of performing bidirectional convolution processing on the sub-combined sequence through a decoder to obtain a recommendation information sequence vector corresponding to the recommendation position includes: The sub-combined sequence is subjected to bidirectional convolution processing by the first decoding layer of the multiple cascaded decoding layers. The convolution result of the first decoding layer is output to the subsequent cascaded decoding layers, so that bidirectional convolution processing and convolution result output can continue in the subsequent cascaded decoding layers until the last decoding layer is output. The convolution result output by the last decoding layer is used as the recommendation information sequence vector corresponding to the recommendation position.

13. An information recommendation device, characterized in that, The device includes: The determination module is used to traverse the historical interaction behavior data of the target object in order to determine the historical interaction information sequence in the historical interaction behavior data. The encoding module is used to perform time-series-based unidirectional encoding processing on the vector of the historical interaction information sequence to obtain the time-series interest vector sequence of the target object; The decoding module is used to concatenate the temporal interest vector sequence and the standard vector to obtain a concatenated vector sequence; and to perform bidirectional decoding on the concatenated vector sequence to obtain a sequence vector of information to be recommended corresponding to each recommendation position, wherein each recommendation position corresponds to the center position of multiple sub-concatenated sequences included in the concatenated vector sequence; The mapping module is used to map multiple sequence vectors of information to be recommended to obtain information to be recommended corresponding to each recommendation position, and to generate a recommendation list based on the information to be recommended corresponding to each recommendation position.

14. The apparatus according to claim 13, characterized in that, The device further includes: The processing module is configured to perform high-dimensional vector encoding on the historical interaction information sequence to obtain a high-dimensional vector corresponding to the historical interaction information sequence; perform low-dimensional vector encoding on the high-dimensional vector to obtain a low-dimensional vector corresponding to the historical interaction information sequence, and use the low-dimensional vector as the vector of the historical interaction information sequence; wherein the dimension of the high-dimensional vector is greater than the dimension of the low-dimensional vector.

15. The apparatus according to claim 13, characterized in that, The encoding module is further configured to divide the historical interaction information sequence based on the temporal sequence of each historical interaction information in the historical interaction information sequence to obtain multiple sub-information sequences included in the historical interaction information sequence; and to perform the following processing on any one of the multiple sub-information sequences: to perform unidirectional convolution processing on the vector of the sub-information sequence through the encoder to obtain the temporal interest vector corresponding to the sub-information sequence; wherein the position of the temporal interest vector corresponds to the edge position of the sub-information sequence; and to combine the multiple temporal interest vectors to use the combination result as the temporal interest vector sequence of the target object.

16. The apparatus according to claim 15, characterized in that, The encoder includes multiple cascaded coding layers, and each coding layer corresponds to a different unidirectional convolution operation; The encoding module is further configured to perform unidirectional convolution processing on the vector of the sub-information sequence through the first encoding layer of the plurality of cascaded encoding layers; The convolution result of the first coding layer is output to the subsequent cascaded coding layers to continue unidirectional convolution processing and output of convolution results in the subsequent cascaded coding layers until the last coding layer is output; the convolution result output by the last coding layer is used as the temporal interest vector corresponding to the sub-information sequence.

17. The apparatus according to claim 13, characterized in that, The decoding module is further configured to perform bidirectional convolution processing on the concatenated vector sequence through the decoder to obtain a sequence vector of information to be recommended corresponding to each of the recommended positions.

18. The apparatus according to claim 17, characterized in that, The decoding module is further configured to divide the spliced ​​vector sequence to obtain the plurality of sub-spliced ​​sequences included in the spliced ​​vector sequence; and to perform the following processing on any one of the plurality of sub-spliced ​​sequences: to perform bidirectional convolution processing on the sub-spliced ​​sequence through the decoder to obtain the recommendation information sequence vector corresponding to the recommendation position.

19. The apparatus according to claim 18, characterized in that, The decoder includes multiple cascaded decoding layers, and each decoding layer corresponds to a different bidirectional convolution operation; The decoding module is further configured to perform bidirectional convolution processing on the sub-concatenated sequence through the first decoding layer of the plurality of cascaded decoding layers; output the convolution result of the first decoding layer to the subsequent cascaded decoding layers, so as to continue bidirectional convolution processing and convolution result output in the subsequent cascaded decoding layers, until the output is reached by the last decoding layer; and use the convolution result output by the last decoding layer as the recommended information sequence vector corresponding to the recommended position.

20. The apparatus according to claim 13, characterized in that, The device further includes: The optimization module is used to perform masking processing on the information to be recommended corresponding to each recommendation position based on the mask, so as to obtain a mask sequence; The decoding module is further configured to perform bidirectional decoding based on the mask sequence to obtain recommendation information sequence vectors corresponding to each recommendation position; and to perform mapping processing on the recommendation information sequence vectors corresponding to each recommendation position to obtain recommendation information corresponding to each recommendation position.

21. The apparatus according to claim 20, characterized in that, The optimization module is further configured to sort the probabilities of each piece of information to be recommended corresponding to the recommendation position in ascending order, determine the information to be recommended corresponding to the first part of the probabilities as the information to be masked, and update the information to be masked among the multiple pieces of information to be recommended to obtain a mask sequence.

22. The apparatus according to claim 20, characterized in that, The decoding module is further configured to combine the temporal interest vector sequence and the mask sequence to obtain a combined sequence; and to perform bidirectional convolution processing on the combined sequence through the decoder to obtain recommendation information sequence vectors corresponding to each recommendation position.

23. The apparatus according to claim 22, characterized in that, The decoding module is further configured to divide the combined sequence to obtain multiple sub-combined sequences included in the combined sequence; and to perform the following processing on any one of the multiple sub-combined sequences: to perform bidirectional convolution processing on the sub-combined sequence through the decoder to obtain a recommendation information sequence vector corresponding to the recommendation position; wherein the recommendation position corresponds to the center position of the sub-combined sequence.

24. The apparatus according to claim 23, characterized in that, The decoder includes multiple cascaded decoding layers, and each decoding layer corresponds to a different bidirectional convolution operation; The decoding module is further configured to perform bidirectional convolution processing on the sub-combined sequence through the first decoding layer of the plurality of cascaded decoding layers; output the convolution result of the first decoding layer to the subsequent cascaded decoding layers, so as to continue bidirectional convolution processing and output convolution results in the subsequent cascaded decoding layers, until the output is reached to the last decoding layer; and use the convolution result output by the last decoding layer as the recommendation information sequence vector corresponding to the recommendation position.

25. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable instructions; A processor, when executing executable instructions stored in the memory, implements the information recommendation method based on artificial intelligence as described in any one of claims 1 to 12.

26. A computer-readable storage medium, characterized in that, It stores executable instructions for inducing a processor to execute, thereby implementing the artificial intelligence-based information recommendation method according to any one of claims 1 to 12.

27. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the information recommendation method based on artificial intelligence as described in any one of claims 1 to 12.

Images